Small-Footprint Keyword Spotting on Raw Audio Data with Sinc-Convolutions